An extremal problem for edge domination insensitive graphs

An extremal problem for edge domination insensitive graphs

Discrete Applied Mathematics 20 (1988) 113-125 North-HolIand 113 Ronald D. DUTTON Department of Computer Science, University of C@mral Florida,Orlan...

1MB Sizes 11 Downloads 116 Views

Discrete Applied Mathematics 20 (1988) 113-125 North-HolIand

113

Ronald D. DUTTON Department of Computer Science, University of C@mral Florida,Orlando, FL 32816, USA Robert C, BRIGHAM Dep~~t~~ntsof Mut~e~~t~csmd Cump~tet Science, ~n~ve~s~ty of Cent~Q~ Florid& Orhzdcr, FL 32816, USA Received 19 October 1984 Revised 4 May 1987 A connected graph is edge domin~ion insensitive if the do~nation number is unchanged when any single edge is removed. The minimum number of edges required by such a graph is determined. Similar results are given when the graph must remain connected upon any edge’s removal and when the dominating set must remain fixed.

1. Introduction All graphs considered are finite and undirected with no loops or multiple edges. An edge joining nodes u and u is indicated by UU.A node u dominates itself and all nodes adjacent to U. A set of nodes D is a dominating set for a graph G if every node of G is dominated by at least one node of D. The domination number y(G) is the size of a smallest dominating set of G. This graphical invariant has been studied extensively [3,4] and critical Grady, where the du~nation number decreases when any vertex is removed [2] or edge added [S], have received recent attention. We shall be concerned with a property which in some sense is the opposite of being edge domination critical. The graph G will be called edge dubitation insensitive if y(G) = Y(G- e) for any edge e of 6;. For brevity we shall say domination insensitive, or even more simply, y-insensitive when the domination number is known to be y‘ This concept corresponds to the y-line-stability number, as defined by Bauer, arary, Nieminen and Suffel [l], being greater than one. The problem to be considered concerns the smallest number of edges required in any y-insensitive graph having p nodes. -VVe will assume throughout that the graph is connected, so yr~V2 [3]. general framework several subproblems are possible, and we consider three, The simplest of the three problems, treated in Section 2, is to determine the minimum number of edges in a graph G with p nodes, domination number y and having the ates aiil 2rcpcrty that some minimum dominating set of G exists which also 0166-218X/88/$3.50

0 1988, Ekevier Science Publishers B.V. (North-Holland)

R. D. Dutton, R.C. Brigham

114

the edge deleted subgraphs G-e. We shall employ &(p~ r) to represent the minimum number of edges in this case. In Section 3 we impose no restriction other than the initial connectedness, and E(pp r) will indicate the minimum number of edges for such graphs. Finally, in Section 4, we treat the case where tne graph must remain connected after any edge is removed. There E,(p, y) ~$1 represent the minimum number of edges. It is interesting to speculate on applications for y-insensitive graphs. One can, for example, contemplate minimum link communication networks having p stations where y of them can transmit a message to the remaining p - y stations with no message traversing more than one communication link. For networks corresponding to y-insensitive graphs this property is preserved whenever a single communication link fails.

Suppose that G is a graph on pi 2 nodes with a minimum dominating set VI= {al,a2, . . 9a,} which also dominates G-e for each edge e of 6. Furthermore, assume that G has the minimum number E&I, v) of edges over all such graphs. We first note thas yz2 since a single node which dominates all other nodes cannot dominate the graph obtained by removing any edge incident to it. Some of the structural properties of such an extremal graph are readily determined. In preparation for the first property let I5 = (b,, b2, . . . , bp_v} be the complement of Vi in V(G). l

Lemma 1. Any extremal graph is bipartite with partite sets VI and V2. Furthermore

each node of a/z has degree 2. Proof. Since VI is the fixed dominating set, dominance is unaffected by edges be-

tween two nodes of VI or two nodes of V2, and thus no such edge is necessary. Clearly any node bi must have degree at least 2 so it can still be dominated by VI when one of its incident edges is removed. On the other hand, it is never necessary to have more than two incident edges since bi can be dominated via any unremoved edge. 0 It follows at once that E&I, JJ)= 2p- 2y whenever extremal graphs exist. We now show there are values of p and y for which there is no y-insensitive graph with a fixed minimum dominating set. For 1sir y define Ai to be those nodes of V2 which are adjacent to ai E VI, i.e., Ai = (bj: aibi IS an edge of G}. emma 2. lAifTAjl* P if&J roof. Suppose lAifTAjl= 1 and let u be the common node* Then {ak: k#i,

j} U {u)

is a dominating set for G since any XE V2- (0) must be dominated by some ak,

An extremal problem for edge domination insensitive graphs

115

k#G, j, because x has degree 2. The size of this new dominating set is y - 1, which is a contradiction. El Lemma 2 implies that lAinAil ==Oor IA;nAil 22. Now assume that G is an extremal graph and let I be the intersection graph of A1,A2, . . . ,A,. The fact that G is connected implies that I also is connected, and it follows that 2 has at least y - 1 edges.

Proof, Consider an edge AiAj of the intersection graph I. Because fAinAji zr:2 there are at least two edges from AiRAi to ai and two more to Qi. Since the nodes of AJ’IAi have degree two, other edges incident to 14113 or Aj in I must correspond to different collections of edges in G, Thus there exists a one-to-one rna~~~~gof the edges of the intersection graph into four element subsets of the edge set of G. Therefore G has at least 4(y- 1) edges. EI Since an extremal graph has 2p-2y ence occurs only when 2p-2y~4y--4,

edges, it follows that the possib~ity of existi.e., p&+-2.

Proof. We construct an extremal graph by specif~ng its edge set as LJ{tzJy i=2y-3,2p-2,...,p-y}.

Fig* 1 iI~ustrates the case when y = 4 and p = 13. Since p z 3 y - 2, each q for 2 s is y is adjacent to at least two nodes of Vz. ft is straightforward to show that VI is a fixed dominating set. Showing that the domination number is y is more difficult but follows by considering minimum do~nating sets which do and do not contain ap q The results of this section are summarized in the following theorem.

a

Fig. 1.

x3

e-w-

‘k-3

‘k

Fig. 2.

TtreOrem1. &(p, y) = 2p - 2y fit ya 2 and pz 3 y - 2, and is u~de~ned otherwise.

Since al! graphs are connected we know that E(p, y)zp - 1. We first treat the special case of y= 1, Theorem 2. E(p, 1)=3p-6

fat pr3.

Proof. Any l-insensitive graph must have three nodes of degree p- 1. Thus E(p, 1)2 3p - 6. Equality follows since a graph having exactly three nodes of degree p- f and no other edges is l-insensitive. Cl We now assume ~12 and consider three cases: pr3y-2, Theorem 3. If ps3y-2,

E(p,y)=p-

pz3y

and p=3y-

1.

1.

Proof. Since E(p, y) zp - 1 we need ollrJ n**r~VA~U~U’C. *nmc*dara tree constructed as illustrated in Fig. 2 with k= 3y-p and I= 3p - 6~. It is not difficult to show that y(G- e) = y(G)=y for all edges e. III For the case pz 3y, we will first determine a count of the minimum number of edges in any graph, not necessarily y-insensitive, for which y 2 2. Arbitrarily select a minimum dominating set DOand let A0 be the set of nodes of G-DO which have exactly one neighbor in DO. We shall show that G has at least 21&j - y edges with an endpoint in & and then use this fact to prove that E(p, y)m2p- 3~. We begin by partitioning DOand ,4* in a parallel manner as fo?lows, Label any

An extremul problem for edge domination insensitive graphs

remaining minimum dominating sets of G by Dr, 02, arbitrary. Define

Ai={oEAo:

N(~)flD~EXi},

. . . , Dn

for dSirn+

117

, where the ordering is

1,

where N(U) is the set of nodes which are adjacent to u. Notice that the collection of sets Xi partitions Do and, since each Ai is the set of nodes in G-i?,, having exactly one neighbor in Xi, the Ai sets partition Ao. Some of the Xi sets, and hence the correspondingAi, may be empty. We shaliEmake use of the fact that the defiition Of Xi means XifIDi=0, for l&92. We shall require another collection of sets Zi defined as follows, for 1s isn: (1) &iDiTt{GEAo: IN(~)n(o,nD,n~==nDi_l)l=l}; (2) Di- Zi dominates G - (Xi U Ai); (3) Zi is maximal with respect to (1) and (2). Thus, for M&n, ZicDiflAo and, for any ZEZi,

Lemma 5. Any node z E Zi is either

(1) in Ai, OF (2) dominates at least one node OfAi which is dmninated by no other node Of

Dia

Proof. Suppose z E Zi- Ai. Since Di - Zi dominates G - (Xi UAi), it follows that the only reason z is in Di is to dominate some node of Xi UAi.But z is in Ao- Ai

and can have no neighbors in Xi. Thus the only possibility is that z dominates at least one node of Ai. 0 An immediate consequence of the above lemma is that we can associate with each z E Zi a unique node Z’EAi, either .z itself if (1) holds or any node dominated by z as described in (2). In the latter case, the selection of z’ from all the nodes of Ai uniquely dominated by z E Di is arbitrary but remains fixed once made. This estanAence blishes a one-to-one corrcs?o,,, n. between the nodes of Zi- Ai and the selected subset Af C_A i 4 Define Bi=(ZifJAi)UAI, Clearly lBiI= IXi I

IZ, I.

for 1s&n.

The next lemma shows that &hesize of these sets is bounded by

l

Lemma 6. IZilS l&l,

for

1 Sian.

roof. The set X# dominates Xi U

dominates

arId ‘i?s

H.D. Dutton, R.C. Brigham

118

I!,n,-Z#=J&[. SinceZ&QandDinXi=O, follows that lZj I s I& 1. Cl

rlloil_lZil+lX~l=Y_1Zil+lXil.

It

We are now ready to count the minimum number of distinct edges which have at least one endpoint in AO. Lemma 7. There are at least 2IAiI- IXil edges with at least one endpoint in Ai, for 1S&n. Proof. For each a EAi there is by definition an edge between a and some node of Xi. This accounts for ]AiI edges. The remainder of the proof establishes a second distinct edge av for each ae Ai- Bi to yield IAi- Bil= IAil - IBil more. Since lBi( = IZiIs IXi 1, the desired count is achieved. The extra edges are defined first for nodes in Al - B1, then for nodes in A2 - Bz and so on through nodes in A,, -B,, . Let a EAi - Bi. There are two possibilities for determining the unique neighbor U. When i= 1, only case (2) applies. (1) a EDj for some j< i. There are two subcases. (1.l) a E2”. By the remarks following Lemma 5, a has a unique neighbor VEBjCAja

(1.2) ae 2” for all j< i.Let j be the smallest index for which a EDimSince a* 2” there is some node v E G - (Xi U Aj) which is uniquely dominsrted by node a in Dj. (2) a@Dj for all j< i. Again there are two subcases. (2.1) a$ Di. Some node v E Di - Xi must dominate a. (2.2) a E Die Since ae Zi it must dominate some node v in G - (Ai UXi) which is dominated by no other node of Di. In each instance, au is a second edge that may be uniquely associated with node a. Edges counted according to (1. l), (2.1) and (2.2) by definition do not involve any node of Xi. Neither does an edge arising from (1.2). Otherwise, node v would be in some Djs for j< i, since Xi= DJIDI f7 0.. flDi_ a- Di, contrary to the definition of v. Thus none of the second edges coincides with any of the original (AiI. Furthermore, no two of the second edges can be the same. The possibility that this can occur for edges arising from (1.1) or (2.2) does not exist since these edges involve a node not in Ai - Bi. The v from an edge found as in (1.2) cannot be in Ai - Bi since it would be dominated by a node x of DjflXi other than a. The v from (2.1) is in Di s SO, if it is also in Ai - Bi, it would have its second edge counted by (2.2) which would make that edge different from au. Thus all edges determined by (1) and (2) are distinct and the total number of such edges is IAil- IBile q The next lemma is crucial to determining an accurate edge count. The edges counted in Lemma 7 for Ai are distinct from the edges counted for Aj, for 1&
An extremal problem for edge domination insensitive graphs

119

Proof. The edges to DOfrom Ai and Ai are certainly distinct since AinAi= 0, and

those arising from (1.1) in the proof of Lemma 7 are distinct by the way o is defined. An edge counted according to (1.2) would not allow v to be in Ak - Dk for k>j since v would be dominated by a node of Dj, in Xk, other than a. If v E Ak -Bk, for k
Recall that this applies to any graph for which ~22. More can be said if the graph is known to be y-insensitive. Lemma 9. If G is y-insensitive there are at least 21A,, + 1I edges with an endpoint in

A n+ 1s distinct fram those counted in Lemma 7. Pwof.

Assume A,, 1 is non-empty since otherwise the result is immediate. Let If aczDj for some j, we can define an edge incident to a as in the proof -4+1* to Lemma 7 and show it is distinct from other edges counted by using the same technique as in the proof to Lemma 8. If a$ Dj, for 1 s jr n, an argument similar to that in the proof to Lemma 8 again shows that a cannot be an endpoint of any edge already counted. Let an be the node of Xn+ 1 to which a is adjacent. Since G a*ed by one of the Dj for some jz 1. In paris y-insensitive, G - aa” must be domin,,, ticular, there is an edge au where v E Dj. We can argue as in the proof to Lemma 7 that if VEA,+~, 2 different edge will be counted for v. Thus there is a;‘:least one distinct edge incident to each a EA,, l which does not involve a node of Xn+ 1. Since there is also an edge from each a E A,, 1 to a node in Xn + 19 we have a total 0 of 21AIP+ 1I previously uncounted edges. Theorem 4. If pz3yr6,

E(p, y)=2p-3~.

roof. From Lemmas 7, 8 and 9, any y-insensitive graph 6’ with yz2 must have at least 21Aol-21A,+I)- ~+IX,+,I~~~A,+,IL~IA,~-~ edges with an endpoint in

R.D. Duttm, R.C. Brigham

120

Fig. 3.

AO. By definition, each node of G- (D&/lo) has at least two edges into De, so neither of these has an endpoint in AO. Thus ECp.y).12/Aol-~+2]G-(DoUlr,)l=21Aol-y+2(p-~-IAol)=2~-3y. We now show, when pz 3 y, that 2p - 3y is an upper bound for E(p, y) by constructing y-insensitive graphs which have this number of edges. If p = 3y, a cycle on 3y nodes is sufficient. Thus assume 2: 3 7 and let t be any positive integer less than y. Consider two cycles Cjt and C3(y_Ij with a node x in one and a node y in the other as shown in Fig. 3. For 15 imp - 3y add a node wi with edges wix and wiy. This graph has p nodes, 2p- 3~9edges, and is y-insensitive. Cl Observe, in Fig. 3, that G -e is connected for every edge e when p - 3~2 2. We will make use of this fact in Section 4. At this point E(p, y) is completely determined except for the case when p = 3y - 1. Since the cycle C+, is y-insensitive, we know that p - 1~ E(3y - 1, y) ,~p. We show that the upper bound is correct. The following lemmas will be useful. The proof to the first is straightforward and the second follows directly from the first. Lemma 10. If G is y-insensitive, no node is adjacent to two or more nodes of degree 1. Lemma 11. If G is a y-insensitive tree, the endpoints of any maximum length path are nodes of degree 1 afdd both are adjacent to a node oyfdegree 2. In the proof for Lemma 12 below, let GX,X2 ...X,denote G - {xI,xz, .*. J,,} for arbitrary nodes x1,x2, . . . ,x,, . .

E(3k-l,k)=3k-1

for integer kz2.

. No tree on five nodes is 2insensitive. Thus, we may assume kz 3. Let G be a tree on the least number of nodes for which E(3k- 1,k)d = 3k- 2. shall show

An extremal problem

for edge domination insensitive graphs

121

no such G exists. By Lemmas 10 and 11 we know there tiarenodes x; y, z of G as chown in Fig. 4(a). No dotted path from x has length greater than 2 and at most one has length 1. We shall examine several cases dependent upon the degree of X. In each case two or three nodes will be removed from G so that the remaining subgraph has y = k- 1, is a tree, and is (k - I)-insensitive. If two nodes are removed the remaining graph hasp = 3(k - 1) and Theorem 4 indicates it cannot be a tree. If three nodes are removed the remaining graph has p = 3(k- 1) - 1 and cannot be a tree because G represents a smallest tree for which p = 3 y - 1. These contradictions for all cases will prove the lemma. Case I: Degree of x= 2. Clearly y(G,,) = k- 1. For e an edge of GX,,zsome minimum dominating set D of G-e includes y. Then D- {y} dominates Gxvt- e and Gwz is our subgraph. Case 2: Degree of x2 4. See Fig. 4(b). The edge wt shown may or may not be present. It is straightforward to see that y(G& = k- 1. For any edge e of GYzsome minimum domination set D of G -e includes y and dominates x by at least one of X, u or w. Thus D- {y} dominates GyT- e and Guzis our subgraph. Case 3: Degree of x= 3 with x leading to two paths of length 2. See Fig. 4(c). Again one easily sees that y(G,,) = k - 1. G - uw can be dominated by a set which includes nodes y and w and a set D’ of size k- 2 which includes X. Then D’U {w} dominates both Gvr- uw, and GYz-xu. For any edge e in GYzt,w,G-e can be dominated by a set D which includes y and v. Thus Guz- e is dominated by D - {y} and G,,* may be taken as our subgraph. Case 4: Degree of x= 3 with x leading to a path of length 1. See Fig. 4(d). Some minimum dominating set of G includes x and y. It follows that y(G,) is either

(a)

(b)

R.D. Dutton, R.C. Brigham

122

k- 1 or k- 2. Suppose first that ~(6,) = k - 1 and e is an edge of G,. Some minimum dominating set D of G - e includes both x and y. Then D - {y} dominates Gwyz- e and G, is our subgraph. Now let y(G,) = k- 2. No minimum dominating set D of G, can include x since DU {y} would then be a k - I domination set of G. We conclude that y(G& = k- 1. GuZ-XXWis dominated by w and a k- 2 domination set of G, and for an edge e in G,, G-e is dominated by a set D’ which includes both x and y so D’- {y} dominates Gvt - e. GuZis thus our subgraph. 0 We summarize the results of this section in the following theorem. Theorem 5 if y=l andpz3, if yz2 andpr3y-2, if yk2 and p=3yif yz2 andpz3y.

We now require that G-e

1,

be connected for every edge e. It is clear that

E,(p, y) ZP and E,(p, y) 2 E(p, y), facts which will be useful below. Theorem6

E,(p,y)=E(p,y),

for y=l,

andforSs3y-lspandp#3y+l.

Proof. The cycle onp nodes proves the theorem for p = 3 y - 1 and 3 y. Otherwise the graphs developed in the proofs of Theorems 2 and 4 show E,(p, y)~E(p, y). Cl Theorem 7. E,(p, y)=3y-2=E@,

y)+ 1, fh-p=3y-2r$.

Proof. Consider the cycle on p nodes.

Cl

Notice that cycles with p< 3 y - 2 do not possess the property that y(G) = y, i.e., y(C,) = [p/31 c y for pc 3y - 2. Thus we may conclude for p in this range that, if such graphs exist, E,(p, y)rp+ 1. Theorem 8. E,(p, y)=E(p, y)+ 1 =p+2, forp=3y+

127.

roof. The graph of Fig. 5 has 3y + 3 = E(p, y) + 1 edges and shows E’Jp, y) sp + 2. Assume E,.(p, y) =p+ 1. Let li be the number of nodes of degree i in an extremal graph. Then 2E,(p,

y) =

21, + 2 =

P-1

i=1

An extrernal probIem for edge domination insensitive graphs

123

Fig. 5.

Since G must be 2 edge-connected, lr =Q. Let /k==p- lj - /d- =*- &_ 1. Thus l

Therefore 2=~s+21’+3~5+=~*+(p-3)1,_1 which implies ri=O for iz5 and either lB= 2 and [4= 0 or l. = 0 and l’= 1. Consider the case when lS= 0 and ld= 1 and observe that every minimum dominating set of G must include the node of degree 4. If one did not, it would contain only degree 2 nodes and could dominate at most 3 y cp nodes. In the proof of Theorem 4 we showed E(p, y)12p - 3y + IXr, I1 where X1+1 is the set of nodes appearing in every minimum dominating set. Therefore we have here that

Now consider the case when I3= 2 and ld=O, illustrated by graph G in Fig. 6. Assume there are m nodes on the left path between nodes x and y, n nodes in the center path and s nodes on the right path, i.e., m + n +s = 3 y - 1. Since the paths can appear in any order we may assume m s n SS. Further we may take m L 1; otherwise x is adjacent to y and G - xy is a cycle on 3 T*+ 1 nodes and requires y + 1 nodes to dominate it. We now show that this graph possesses an edge e for which y(G -e) = y(G) + 1. Then E,(3 y + 1, y) > 3 y + 2 and is therefore 3 y + 3. Consider re&movinganv edee to x. The resulting graphs are all similar in the sense that - -69-incident ---they are a c&e with a pendant path and the analysis which follows applies to any one of them. Assume u is adjacent to x on the left path and let G’= G- vx. By assumption y(G’) = y(G) = y and 6;’ consists of a cycjie on n +s + 2 = 3 y + 1 -m nodes with a pendant tail of length EZ.We discuss three cases dependent upon the value of m. X

Fig. 6.

RD. Dutton, R.C. Brigham

124

Case 1: m = 3 k-forsome k 11. Every minimum dominating set contains k or k + 1 nodes from the pendant path. In the former case at least y - k + 1 more nodes are needed to dominate the cycle and in the latter case at least y-k more. In any event we require a total of y + 1 nodes. Therefore m # 3k, nor is n or s equal to 3k since a similar argument holds for them. Case 2: m = 3k+ 1 for some kz0. Select y - k cy& nodes, including y, to dominate the cycle and one path node. Now select k path nodes to dominate the other 3k nodes. Thus it is possible for any of m, n, or s to be 3k+ 1 for some kz0. Case 3: m = 3 k + 2 for some kr 8. The path nodes and y can be dominated by k+ 1 path nodes. Then we still will need at least y - k cycle nodes to complete a minimum dominating set, i.e., y + 1 total. Therefore neither m, n nor s may be 3k+2 for any krU. Thus m, n and s must each satisfy Case 2 and have values of 3kl + 1, 3kz+ 1 and 3k3+ 1, respectively, for non-negative integers kl, kz, and kS. But this implies q p=m+n+s+2=3(kl+kz+k3+ 1)+2+3y+ 1. Therefore &,(p,y)>3y+2. We summarize the results of this section in the fo~lo~ng theorem. Theorem 9 r E@y)=3p-6 E&9 39=

E(p,y)+l=p+2

if y=l atidp13,

if y12 andp=3y-2, if yr2 andp=3y-1

or 3y,

ify&andp=3y+l, v yr2 andpz3y+2.

The situation when pi 3 y - 3 is unknown. We have been able to show that no graphs exist which meet the stated requirements when pi 3y- 3 and y =2,3 or 4. However, Fig. 7 shows there is, for example, a &insensitive graph with p = 3 y - 3 which remains connected when any edge is removed.

An extremal problem for edge domination insensitive graphs

125

feremes ill D. Bauer, F. Harary, J. Niemiuen and C. Suffel, Domination alteration sets in graphs, Discrete Math. 47 (1983) 153-161.

PI R.C. Brigham, P.Z. Chinn and RD. Dutton, Vertex domination-critical graphs Networks, to appear. 131E.J. Cockayne, Domination of undirected graphs - a survey, in: Y. Alavi and D.R. Lick, eds., Theory and Applications of Graphs in America’s Bicentennial Year (Springer, New York, 1976) 141-147. 141R. Laskar and H.B. Walikar, On domination related concepts in graph theory, Lecture Notes in Math. 885 (Springer, New York, 1980) 308-320. D. Summer and P. Blitch, Domination critical graphs, J. Combin. Theory (B) 34 (1983) 65-76. PI